FE 445 – Investment Analysis and Portfolio
Management
Fall 2020
Farzad Saidi
Boston University | Questrom School of Business
Lectures 3 & 4: Performance of
securities
Ex-post past returns
Holding-period return (HPR): one-period investment, regardless of the
length
P1 + CF − P0 P1 + CF
HPR = or HPR = −1
P0 P0
• P1 : Ending price (or sale price)
• P0 : Beginning price (or buy price, $ you put up at 0)
• CF : Cash flow during holding period (e.g., dividend or interest
payment)
Assumption: cash flows are received at the end of the period (important
if holding period is long)
1
Arithmetic average
Finding the average HPR for a time-series of returns:
n
X HPRt
HPRavg = , n = # time periods
t=1
n
Example: What is the average return?
2011 -21.56%
2012 44.63%
2013 23.35%
2014 20.98%
2015 3.11%
2016 34.46%
2017 17.62%
2
Arithmetic average
Finding the average HPR for a time-series of returns:
n
X HPRt
HPRavg = , n = # time periods
t=1
n
Example: What is the average return?
2011 -21.56%
2012 44.63%
2013 23.35%
2014 20.98%
2015 3.11%
2016 34.46%
2017 17.62%
−0.2156 + 0.4463 + 0.2335 + 0.2098 + 0.0311 + 0.3446 + 0.1762
HPRavg = = 17.51%
7
However, arithmetic average ignores compounding!
2
Geometric average
Geometric average takes compounding into account:
n
!1/n
Y
HPRavg = (1 + HPRt ) −1
t=1
HPRavg = (0.7844 × 1.4463 × 1.2335 × 1.2098 × 1.0311 × 1.3446 × 1.1762)1/7 − 1
= 15.61%
This is the fixed return that would yield the same after 7 years.
3
Arithmetic vs. geometric average
year value
2015 $100
2016 $50
2017 $75
−0.5+0.5
• Arithmetic: 2 =0
• Geometric: (0.5 × 1.5)1/2 − 1 = 0.866 − 1 = −0.134
4
Arithmetic or geometric: which one to use?
When you are evaluating past returns (ex post):
• Use arithmetic average if you are not reinvesting, e.g., have $100
portfolio in the beginning of every year
• Use geometric average if you are reinvesting everything
When trying to estimate an expected return (ex ante):
• Use arithmetic average: average of one-period returns
Note: The difference between arithmetic and geometric mean is due to
the variance of returns.
5
Annualizing returns
Holding period of less than one year (n periods per year)
• Without compounding (annualized percentage rate):
APR = HPR × n
• With compounding (effective annual return):
n
EAR = (1 + HPR) − 1
Holding period of more than one year (T years)
• Without compounding (annualized percentage rate):
APR = HPR/T
• With compounding (effective annual return):
1/T
EAR = (1 + HPR) −1
6
Problem: APR and EAR
You bought a 3-month T-bill (discount bond) with face value of $100 for
$95.23. What is the annual rate of return with and without
compounding?
• The 3-month holding period return is:
HPR = 100/95.23 − 1 = 5.0%
• Without compounding: APR = 5.0% × 4 = 20.0%
• With compounding (e.g., because you reinvest):
EAR = (1.05)4 − 1 = 21.6%
7
Dollar-weighted rate of return
How can one assess the profitability of a potential project/investment?
⇒ Find the internal rate of return for a series of cash flows:
CF0 CF1 CF2
$0 = 0 + 1 + 2 + ···
(1 + IRR) (1 + IRR) (1 + IRR)
IRR: discount rate that sets the NPV of CFs equal to zero
• Higher IRR ⇒ more desirable investment
• One of many metrics
8
Example
year beginning-of-year price dividend paid at year-end
2010 $100 $4
2011 $110 $4
2012 $90 $4
2013 $95 $4
Scenario: An investor buys three shares of the above stock at the
beginning of 2010, buys another two shares at the beginning of 2011,
sells one share at the beginning of 2012, and sells all four remaining
shares at the beginning of 2013.
1. What are the arithmetic and geometric average time-weighted rates
of return for the investor?
2. What is the dollar-weighted rate of return?
9
Solution Part I
• Arithmetic average:
((110–100+4)/100+(90–110+4)/110+(95–90+4)/90)/3 = 3.15%
• Geometric average:
(1.14 × 0.8545 × 1.1)(1/3) − 1 = 2.33%
10
Solution Part II
• Calculate the IRR for the following four cash flows:
• Beginning of 2010: 3 × (−100) = −300
• Beginning of 2011: 3 × 4 − 2 × 110 = −208
• Beginning of 2012: 5 × 4 + 90 = 110
• Beginning of 2013: 4 × (4 + 95) = 396
11
Solution Part II
• Calculate the IRR for the following four cash flows:
• Beginning of 2010: 3 × (−100) = −300
• Beginning of 2011: 3 × 4 − 2 × 110 = −208
• Beginning of 2012: 5 × 4 + 90 = 110
• Beginning of 2013: 4 × (4 + 95) = 396
• IRR = –0.1661%
11
Real interest rates
• Interest rate that is adjusted for (expected) changes in the price
level
1. Ex-ante real rate of interest based on expected level of inflation
2. Ex-post real rate of interest based on observed/realized level of
inflation
• Real interest rate more accurately reflects true cost of borrowing
• When the real rate is low, there are greater incentives to borrow and
less incentives to lend
12
Real vs. nominal interest rates
Approximation of Fisher equation:
rreal ≈ rnom − i,
where rreal (rnom ) is the real (nominal) interest rate and i is the inflation
rate.
Exact Fisher equation is:
1 + rnom
1 + rreal = or
1+i
rnom − i
rreal = .
1+i
Example:
rnom = 9%, i = 6%,
rreal = 3% or
rreal = 3%/1.06 = 2.83%.
13
Realized real returns
14
Negative interest rates
• In November 1998, rates on Japanese 6-month government bonds
were negative
• Investors were willing to pay more than they would receive in the
future
• Same thing happened in the U.S. in September of 2008, then
Sweden (July 2009), Denmark (July 2012), the Eurozone (June
2014), Switzerland (December 2014), and Japan again in early 2016
• Negative rates best justified by “convenience fee” for not having to
hold cash
• Enormous role for negative nominal rates rather than negative real
rates ⇒ money illusion (as real rates have been negative for a much
longer period of time)
15
Expected vs. realized returns
• Expected return (E (r )): the mean of the return based on a
mathematical model or simply a guess
• Realized (historical) return (¯
r ): returns you observed in the past
Example: Fair dice
• with equal probabilities: 1, 2, 3, 4, 5, 6
• 10 rolls: 1, 2, 5, 1, 3, 6, 5, 6, 4, 1
Averages:
• Realized average: r¯ = 3.4
• With an infinite number of rolls, one approaches the expected mean:
E (r ) = 3.5
16
Expected mean and variance
Expected returns: X
E (r ) = ps rs
s
Variance of returns:
X 2
Var (r ) = σ 2 = ps × (r − E (r ))
s
• s: states of the world (scenarios): 1, 2, . . . , S
• ps : probability of state s
P
• Probabilities sum up to one: s ps = 1
• rs : return if a state occurs
√
• σ = σ 2 : standard deviation (or volatility)
17
Problem: scenario analysis
state scenario prob return
1 Recession 0.2 -0.05
2 Normal 0.5 0.05
3 Boom 0.3 0.15
P
• E (r ) = s ps rs
2
P 2
• Var (r ) = σ = s ps (rs − E (r ))
σ = 0.07, comparable to return deviations
18
Ex-post average return and variance
Using n periods of historical data
Average:
n
X
r¯ = rt /n
t=1
Variance:
n
1 X 2
σ̄ 2 = (rt − r¯)
n − 1 t=1
• rt is the HPR in period t
• The denominator in the variance is n − 1 to adjust for the fact that
the mean used in the formula is an estimate (degree of freedom)
• What is the ex-post variance of the 10 dice rolls?
19
How to yield expected returns?
Use past (ex-post, historical) data
• Use arithmetic (simple) average to yield expected return
• It delivers objective numbers: used a lot
• Usually a pretty good bet
• Problem: the world changes
• And so does the covariance of prices/returns
• For instance, rating agencies evaluated mortgage-backed securities,
neglecting substantial correlations
Use subjective ad-hoc forecasts
• Problem: anything goes!
20
How could we resolve this?
state return
good 15%
bad -10%
If you know the probabilities (pg = 0.7 and pb = 0.3), then
21
How could we resolve this?
state return
good 15%
bad -10%
If you know the probabilities (pg = 0.7 and pb = 0.3), then
E (r ) = 0.7 × 0.15 + 0.3 × (−0.10) = 7.5%
If you don’t know the probabilities
• In the past 100 years there were 71 good and 29 bad years
• Guess: pg = 71/100 = 0.71 and pb = 29/100 = 0.29
E (r ) = 0.71 × 0.15 + 0.29 × (−0.10) = 7.75%
100
X 71 × 0.15 + 29 × (−0.10)
r¯ = rt /n = = 7.75%
t=1
100
21
Stock index: S&P 500
• Value-weighted (e.g., S&P 500) vs. price-weighted (e.g., DJIA)
22
Stock index: S&P 500
• Value-weighted (e.g., S&P 500) vs. price-weighted (e.g., DJIA)
• Broad-based index vs. sector indices
22
Stock index: S&P 500
• Value-weighted (e.g., S&P 500) vs. price-weighted (e.g., DJIA)
• Broad-based index vs. sector indices
• S&P 500
• “Widely regarded as the best single gauge of the U.S. equities
market, this world-renowned index includes 500 leading companies in
leading industries of the U.S. economy.” (Standard & Poor’s website)
• Represents approximately 75% of U.S. market cap
• Important benchmark for portfolio managers
• Does not include dividends
22
S&P 500 1950–2019: index level
3000
2500
2000
S&P500 price level
1500
1000
500
0
1950 1960 1970 1980 1990 2000 2010 2020
23
S&P 500 1950–2019: log returns
0.2
0.15
0.1
0.05
S&P500 return
-0.05
-0.1
-0.15
-0.2
-0.25
1950 1960 1970 1980 1990 2000 2010 2020
Simple (normal) returns: Rt = (Pt − Pt−1 )/Pt−1
Continuously compounded (log) returns: rt = log(Pt /Pt−1 )
24
S&P 500 1950–2019: histogram
80
70
60
50
Normal distribution
40
30
20
10
0
-0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2
25
S&P 500 1950–2019: summary statistics
Monthly log returns 01/01/1950–01/01/2019
mean 0.609% (7.31% annualized)
std 4.135% (14.32% annualized)
min -24.543% (1 Sept 1987)
max 15.104% (1 Sept 1974)
skewness -0.668
kurtosis 5.464
autocorr 0.042
26
Volatility as a measure of risk
• If financial returns are IID (independent and identically distributed)
normal
• then the first two moments completely describe the distribution
27
Volatility as a measure of risk
• If financial returns are IID (independent and identically distributed)
normal
• then the first two moments completely describe the distribution
• ...and volatility is the measure of risk
• Alternatively, for an agent with mean and variance preferences,
volatility is also a sufficient measure of risk
Unfortunately, this is not necessarily true...
27
Kurtosis and skewness
• Skewness measures the degree of asymmetry of a distribution
⇒ Zero skewness corresponds to a symmetric distribution
⇒ Positive skewness indicates a relatively long right tail
⇒ Negative skewness indicates a relatively long left tail
n 3
1 X rt − r¯
skewness =
n t=1 σ̄
• Kurtosis measures the extent to which probability is concentrated in
the center and tails of a distribution
⇒ High values of kurtosis indicate heavy tails, and low values indicate
light tails
n 4
1 X rt − r¯
kurtosis =
n t=1 σ̄
28
Value-at-Risk (VaR)
• Definition: p VaR is defined such that the probability of a loss
greater than VaR is (at most) p, while the probability of a loss less
than VaR is (at least) 1 − p
• Mathematically (use minus sign because VaR is a positive number):
Z −VaR(p)
Pr[q ≤ −VaR(p)] = p = f (q)dq
−∞
⇒ Negative skewness, prominent in stock returns, is a source of
concern for VaR
29
Example: VaR
2 assets:
• A, B, independent and with same distribution
• 4.9% chance return is −100
• 95.1% chance return is 0
What is the Value-at-Risk of assets A and B?
30
Example: VaR
2 assets:
• A, B, independent and with same distribution
• 4.9% chance return is −100
• 95.1% chance return is 0
What is the Value-at-Risk of assets A and B?
• VaR(5%) = 0, VaR(1%) = 100
What is the Value-at-Risk of a portfolio consisting of A and B?
30
Example: VaR
2 assets:
• A, B, independent and with same distribution
• 4.9% chance return is −100
• 95.1% chance return is 0
What is the Value-at-Risk of assets A and B?
• VaR(5%) = 0, VaR(1%) = 100
What is the Value-at-Risk of a portfolio consisting of A and B?
• Equally weighted portfolio with A and B:
VaR(1%)(0.5A + 0.5B) = 50 < 100 = 0.5 × VaR(1%)(A)
+ 0.5 × VaR(1%)(B)
30
Value-at-Risk inputs
• Probability (p)
31
Value-at-Risk inputs
• Probability (p)
• Most common level is 1% (equivalent: confidence level c = 99%)
31
Value-at-Risk inputs
• Probability (p)
• Most common level is 1% (equivalent: confidence level c = 99%)
• Holding period
31
Value-at-Risk inputs
• Probability (p)
• Most common level is 1% (equivalent: confidence level c = 99%)
• Holding period
• Time period over which losses can occur
• Must be long enough for corrective measures and needs to reflect
illiquidity of assets
• Usually one day
• Longer holding periods more realistic for institutional investors,
shorter (intra-day) VaR for trading desks
31
Value-at-Risk inputs
• Probability (p)
• Most common level is 1% (equivalent: confidence level c = 99%)
• Holding period
• Time period over which losses can occur
• Must be long enough for corrective measures and needs to reflect
illiquidity of assets
• Usually one day
• Longer holding periods more realistic for institutional investors,
shorter (intra-day) VaR for trading desks
• Probability distribution
31
Value-at-Risk inputs
• Probability (p)
• Most common level is 1% (equivalent: confidence level c = 99%)
• Holding period
• Time period over which losses can occur
• Must be long enough for corrective measures and needs to reflect
illiquidity of assets
• Usually one day
• Longer holding periods more realistic for institutional investors,
shorter (intra-day) VaR for trading desks
• Probability distribution
• Estimated using past observations and statistical model
31
Interpreting VaR
• VaR numbers are meaningless without probability and holding period
• Capital required against failure and for internal controls
• VaR violation occurs when trading losses exceed VaR:
• For 1% daily VaR we expect to lose less than VaR 99 days out of 100
• With 250 trading days per year, the expected number of violations is
then 2.5 per year
32
Interpreting VaR
• VaR numbers are meaningless without probability and holding period
• Capital required against failure and for internal controls
• VaR violation occurs when trading losses exceed VaR:
• For 1% daily VaR we expect to lose less than VaR 99 days out of 100
• With 250 trading days per year, the expected number of violations is
then 2.5 per year
• Facilitates comparisons between asset classes, risks summarized in a
single number
32
Interpreting VaR
• VaR numbers are meaningless without probability and holding period
• Capital required against failure and for internal controls
• VaR violation occurs when trading losses exceed VaR:
• For 1% daily VaR we expect to lose less than VaR 99 days out of 100
• With 250 trading days per year, the expected number of violations is
then 2.5 per year
• Facilitates comparisons between asset classes, risks summarized in a
single number
• Is VaR positive or negative?
• VaR can be presented as positive or negative number
• Probabilities can be stated as close to one or close to zero
(VaR(99%) or VaR(1%))
32
Capital requirements of banks
• Required bank capital against market risk are based on VaR (since
1996 amendment to 1988 Basel I)
• Inputs:
• Probability: 1%
• Holding period: 10 days or two calendar weeks
• Observation period based on at least a year of data, updated at least
once a quarter
• Regulatory framework allows capital to depend on accuracy of VaR
model ⇒ can have important consequences if the VaR model is
defective
• Capital is higher of previous day’s VaR and average VaR over last 60
business days times a “multiplier” k ≥ 3
• Higher k imposed by regulators if backtesting reveals large number
of exceptions
• Required capital against credit and operational risk is based on a
one-year VaR with 99.9% confidence level (Basel II)
33
Capital requirements in the insurance industry
• Capital requirements for most fixed-income assets in the insurance
industry are tied to credit ratings ⇒ thought to be a function of
VaR, covering unexpected losses
• Since 2009 (2010) capital requirements for RMBS (CMBS) are no
longer based on ratings
• Proprietary risk measure purchased from Pimco (RMBS
2009 − 2015) and BlackRock (RMBS 2016−, CMBS 2010−)
• For each security, expected loss (“ELOSS”) is calculated annually
• Expected loss of principal, discounted at coupon yield
• ELOSS ∈ [0, 1] ⇒ IP (intrinsic price) = 1 − ELOSS
BV −IV
• BV determines NAIC 1 − 6 (if ELOSS = 0 ⇒ NAIC 1)
• where IV (intrinsic value) = IP × Par value
34
New capital requirements do not provide buffer against unexpected losses
35
New capital requirements do not provide buffer against unexpected losses
Capital requirements = RBC% × BV ≈ BV − IV
35
New capital requirements do not provide buffer against unexpected losses
Capital requirements = RBC% × BV ≈ BV − IV
35
New capital requirements do not provide buffer against unexpected losses
Capital requirements = RBC% × BV ≈ BV − IV
35
Summary
This class:
• Expected vs. realized returns
• Use arithmetic mean of past (realized) returns to get E (r )
• Distribution of returns
• σ and VaR a good measure of risk for normal distributions
Next class:
• The risk-return trade-off
• Portfolios
36
